To solve the unrelated parallel machine scheduling problem (UPMSP) with preventive maintenance (PM) and sequence dependent setup time (SDST), a multi-colony artificial bee colony (MABC) algorithm is proposed to minimize makespan and total tardiness simultaneously. In this algorithm, employed bees are divided into s colonies. Except for the worst employed bee colony, each employed bee colony corresponds to a onlooker bee colony. Combined with the characteristics of two objective functions, PM and SDST, three kinds of neighborhood searches are designed. Different combinations of global search and neighborhood search are used to implement the employed bee phase and the onlooker bee phase, and two elimination processes are applied. Experimental research on the strategy and search performance of the MABC is carried out, and computational results demonstrate the effectiveness of the proposed strategy and the search advantage of the MABC.